ECMWF Seminar Proceedings
Land surface versus ocean influence
on atmospheric variability and predictability
at the seasonal timescale
H. Douville, B. Decharme, S. Conil, Y. Peings
Corresponding author address:
Dr. Herv Douville
CNRM/GMGEC/UDC, 42 avenue Coriolis
31057 Toulouse Cedex 01, France
Although the chaotic nature of the atmospheric dynamics (that is its sensitivity to
initial conditions) is a fundamental limit to its deterministic predictability, it is now
well known that predictability of seasonal weather statistics is possible (see reviews
by Palmer and Anderson 1994). This arises from what may be termed external
factors that alter the likelihood of residence in atmospheric attractors, enabling
probabilistic forecasts to be made of the seasonal mean state, on condition that the
external forcing is itself predictable.
It is commonly accepted that the primary source of such external forcing at
seasonal timescales arises from anomalous sea surface temperature (SST) patterns
(i.e. Rowell 1998). On the one hand, these can indeed be predicted, either using
coupled dynamical models or statistical models. On the other hand, both observational
and numerical studies suggest that interannual atmospheric variability is partly driven
by SST variability, especially in the tropical Pacific given the major teleconnections
associated with the El Nio Southern Oscillation (i.e. Wallace et al. 1998).
Further potential sources of predictive skill, in particular land surface anomalies,
are generally believed to be much less important and are often neglected in studies
about seasonal predictability. Nevertheless, an increasing body of literature suggests
that land surface memory can also contribute to atmospheric variability and
predictability at the monthly to seasonal timescale. The main objective of this paper is
to relate a brief history of this research activity, to summarize some important results
and to raise the issues to be further explored. It is not intended to be a comprehensive
review of the subject, but rather to illustrate both advances and issues in the field,
mainly using some of the studies conducted over the last few years at CNRM.
2. Land surface models and data
Land surface models (LSM) have evolved substantially from the original bucket
model of Manabe (1969). While they were initially developed as simple
parametrizations within the atmospheric general circulation models (GCM), their
increasing complexity and the need of validating them in off-line mode (i.e. driven
by more realistic atmospheric forcings than those simulated by climate models) has
led to a progressive emancipation of LSMs that have moved from being subroutines
to independent models which can be coupled to atmospheric models but also used
off-line for other applications such as validation-calibration, but also hydrological
predictions or impact studies (Polcher et al. 1998).
The status of LSMs is therefore getting closer to that of oceanic GCMs, but does
it mean that their influence on atmospheric variability is of comparable relevance?
One crucial issue for answering this question is the lack of observational datasets to
document the land surface variability at the global scale. While in situ and - since the
1980s - satellite observations do exist to characterize the large-scale monthly SST
variability over the 20th
century, the task is much more difficult over land for at least
three reasons. First, the land influence on atmosphere depends on several parameters,
not only surface temperature but also hydrology, vegetation and topography. Second,
some of these parameters, especially but not only the subsurface hydrology, are still
difficult to retrieve from satellite observations. Third, land surfaces generally show a
much stronger high-frequency variability as well as a much stronger spatial
heterogeneity than ocean surfaces, so that the monitoring of land surface variability
with in situ observations only is simply not possible.
Let us here examine the modeling implications of this last remark given the focus
of the ECMWF seminar on the parametrization of subgrid physical processes. One of
the main challenge in developing models for the land surface hydrology in numerical
climate models stems indeed from the fact that the horizontal resolution of these
models is incompatible with the characteristic scales of surface hydrologic properties.
What cannot be explicitly resolved within the numerical model discretization must be
parameterized. While in the 1980s, the development and local validation of LSMs has
focused on improving the one-dimensional structure of the soil-snow-canopy system,
more attention has been paid recently to the treatment of the horizontal subgrid
variability of (i.e. Entekhabi and Eagleson 1989, Dmenil and Todini 1992, Koster
and Suarez 1992).
Fig. 1: The main sources of subgrid hydrological variability acoounted for in the ISBA-SGH land surface model of CNRM (Decharme and Douville 2006): precipitation, orography and vegetation.
Beyond numerical sensitivity studies showing the relevance of land surface
heterogeneities for the calculation of the land surface energy and water fluxes, basin-
scale field experiments have been a key stage in the development, calibration and
validation of subgrid hydrological processes (i.e. Boone et al. 2004). River discharge
observations indeed offer the opportunity to validate the simulated runoff (not all the
components of the water budget) after integration over the drainage area. Depending
on the size of the basin and on the frequency of interest, river routing models might be
however necessary to account for the time lag between the production of runoff and
the discharge response at the gauging station.
At CNRM, a coherent subgrid hydrology, ISBA-SGH, accounting for grid cell
heterogeneities not only in soil and vegetation properties (tile or mosaic approach,
Koster et al. 1992) but also in precipitation intensities (Entekhabi and Eagleson 1989)
and orography (Topmodel approach, Beven and Kerby 1979) has been developed and
tested over the French Rhne river basin (Decharme and Douville 2006). Off-line
simulations at 8km driven by the SAFRAN meteorological analysis of CNRM have
been compared with lower resolution (1) simulations after linear interpolation of the
high-resolution atmospheric forcing. For drainage areas exceeding a few 1 by 1 grid
cells, results showed that the daily river discharge efficiencies of ISBA-SGH are not
only less sensitive to horizontal resolution than those of the standard ISBA model, but
also exceed at low resolution those obtained with ISBA at high resolution (Fig. 2).
Fig. 2: Cumulative efficiency (Nash criterion) distribution of daily river discharges simulated at 88 gauge stations distributed over the Rhne river basin. All hydrological simulations are driven by observed atmospheric forcings either at a 8km (solid lines) or 1 (dashed lines) horizontal resolution and coupled to the MODCOU river routing model. ISBA-SGH (in red) is compared to the standard ISBA model (in black). From Decharme and Douville 2006.
Beyond basin-scale simulations, there is a need of validating and/or comparing
LSMs at the global scale at which they are used in atmospheric GCMs. This was the
purpose of the International Satellite Land Surface Climatology Project (ISLSCP) that
has been launched in the mid-1990s and has allowed land surface modelers to produce
global soil moisture climatologies by driving their models with common atmospheric
forcings and land surface parameters in the framework of the Global Soil Wetness
Project (GSWP, http://grads.iges.org/gswp).
The ISBA land surface model of CNRM contributed to GSWP and was driven by
the ISLSCP atmospheric forcings first from 1987 to 1988 (GSWP-1, Douville 1998),
then from 1986 to 1995 (GSWP-2, Decharme and Douville 2007). Besides control
runs using the common ISLSCP soil and vegetation parameters, parallel integrations
have been achieved with the native ISBA land surface parameters to produce soil
moisture and snow mass climatologies that are fully consistent with the CNRM
atmospheric GCM. Such climatologies can be used to nudge global atmospheric
simulations towards realistic land surface boundary conditions and compare the
influence of soil moisture or snow mass versus SST on atmospheric variability and
predictability at the seasonal timescale (see next section).
Fig. 3: Cumulative efficiency (Nash criterion) distribution of monthly river discharge simulated from 1986 to 1995 at 80 gauge stations distributed over the largest (>100000 km) worlds river basins. Six land surface models are compared : ISBA-SGH (NEW) and ISBA-standard (dt92), as well as four other LSMs having contributed to the international GSWP-2 intercomparison project (B3 atmospheric forcing). Runoff from all simulations has been converted into river discharge using the TRIP river routing model (Oki and Sud 1998). From Decharme and Douville 2007.
Moreover, the off-line GSWP simulations were also useful to test the ISBA-SGH
hydrology at th